16 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
weaselbot	e02dd41f74	Address review: drop unused bti c landing pads, fix rationale comment	2026-08-03 09:17:28 -04:00
weaselbot	d040ed63a4	Move aarch64 Node16 SIMD index loads into assembly ... The aarch64 NEON paths in getNodeIndex/getNodeIndexExists, getChildGeq(Node16*), scan16, and checkMaxBetweenExclusiveImpl<Node16> loaded the full 16-element Node16::index array (and the Node48 reverseIndex array via scan16) with NEON intrinsics, then masked the result down to [0, numChildren). Only the in-use slots are initialized; the unused bytes are indeterminate, so the wide loads were undefined behavior in C++ (per [basic.indet]) even though the trailing lanes were discarded. MSan reports this on x86-64; on aarch64 it is the same UB but MSan's imprecise modeling doesn't flag it at -O0, so there is no red->green test. Mirror the existing x86-64 fix (commit 6fed133): implement the index operations in file-level assembly, where loading and operating on indeterminate values is well-defined. Add simd_aarch64.S with find_eq_16, find_ge_16, and mask_in_range_16. AArch64 lacks pmovmskb, so (like the prior NEON code) these return a 64-bit nibble mask rather than a 16-bit bitmask; the C++ call sites keep their existing nibble-mask arithmetic and only swap the inline NEON load/compare for the assembly call. The childMaxVersion compares stay in C++ NEON intrinsics, matching x86-64's compare16: those slots are always initialized to zero by the allocator, so the wide loads are defined. The assembly functions carry `bti c` landing pads and the same aeabi_feature_and_bits attributes the compiler emits for -mbranch-protection=standard, so the object stays BTI/PAC/GCS-compatible (and warning-free under -z force-bti). CMakeLists.txt builds simd_aarch64.S into the object library and the SIMD test/bench/fuzz targets on aarch64. Closes #68	2026-08-02 23:48:28 -04:00
weaselbot	6b1f597d77	Fix ARM NEON scan helpers truncating 64-bit versions ... When USE_64_BIT=1, InternalVersionT stores an int64_t, but the aarch64 NEON fast paths in scan16 (both the indexed Node48 variant and the Node256 variant) and checkMaxBetweenExclusiveImpl(Node16) only copied the low 32 bits of each version into uint32x4_t lanes and compared against a truncated 32-bit readVersion. This produced wrong conflict/commit decisions once any version exceeded 2^32 - 1. Extract the per-version "greater than readVersion" computation into a shared conflictMask16 helper. For 64-bit versions it uses vcgtq_s64 on the full int64 values and narrows the resulting 16-byte mask into the same nibble-packed layout the callers already consume. The 32-bit path is preserved unchanged. Closes #69	2026-08-02 23:35:32 -04:00
weaselbot	70630649d9	ci: run MSan on arm64 ... Add a debug-arm64 matrix entry to the test job that runs the MSan fuzz tests on an arm64 runner. The arm instrumented libc++ toolchain (22.1.8) is built against LLVM 22, so this entry installs clang-22 from apt.llvm.org while the existing amd64 entries keep clang-21. Generalize the test job matrix with runner/arch/llvm_version/msan_url fields so the runner, apt cache key, LLVM toolchain version, and MSan toolchain download URL are selected per matrix entry. Closes #73	2026-08-02 23:28:40 -04:00
weaselbotandandrew	0b12a037c4	Correct rationale in simd_x86_64.S comment ... The SIMD operations are written in assembly because loading and operating on indeterminate values is UB in C++ but well-defined in assembly. msan not tracking the loads is a side effect, not the reason. Closes #71	2026-08-03 02:37:18 +00:00
weaselbot	2231c093df	Fix null pointers passed to memcmp/memcpy in skip_list ... In addWrites, the sizing constructor std::vector<KeyInfo>(count * 2) was used instead of reserve(count * 2), leaving count*2 default- constructed (null/empty key) entries that were then compared via operator<, passing nullptr to memcmp. Switch to reserve so only real entries exist. Additionally guard the memcmp/memcpy calls in operator<, SkipList::less, and copyToArena against empty spans (where data() may be nullptr), which is reachable on the first setOldestVersion with an empty removal key. Closes #64	2026-07-13 12:04:22 -04:00
weaselbot	dee3a8f640	Return 0 instead of -1 from hash_table getBytes() ... The hash_table implementation cannot accurately track memory usage and was returning -1 from ConflictSet::getBytes() and ConflictSet_getBytes(). That violates the API contract that getBytes() returns a non-negative byte count. Change both entry points to return 0 and document in ConflictSet.h that implementations which do not track memory usage may return 0. Add a regression test in test_conflict_set.py that loads the hash_table implementation and verifies getBytes() is non-negative. Closes #62	2026-07-07 11:00:52 -04:00
weaselbot	4fcdc5d7e9	Fix RealDataBench subspan and version API contract violations ... - Use subspan count `line.size() - 2` instead of `line.size()` to avoid reading past the line bounds, and guard lines shorter than the two-byte prefix. - Clamp `readVersion` and `setOldestVersion` arguments to `0` so the conflict set never receives negative, non-monotonic versions during the warmup phase. Fixes #58.	2026-06-29 14:01:23 -04:00
weaselbot	63f9a139da	Fix undefined behavior on empty input in strinc() and prefixRange() ... strinc() in ConflictSet.cpp used std::string_view::size() (size_t) and subtracted 1 without first checking for an empty string. For the root node, getSearchPath() returns the empty string, so every debug correctness check underflowed size_t and relied on implementation-defined conversion to signed int. prefixRange() in Bench.cpp had the same loop shape. Although TrivialSpan::size() returns int, on an empty (or all-0xff) key the function then asserted and continued executing, allocating a zero-length buffer and writing before its start. Changes: - In strinc(), initialize index as signed int(str.size()) - 1 so the loop is skipped for empty input, and return ok=false cleanly. - In prefixRange(), initialize index the same way and call std::abort() after the assert so invalid input cannot fall through to an out-of-bounds write. - Replace C-style uint8_t casts with explicit static_casts. Closes #59	2026-06-29 13:52:02 -04:00
weaselbotandandrew	971deb477c	Keep Python key buffers alive in WriteRange/ReadRange ... `write()` and `read()` created _Key objects from ephemeral ctypes arrays backed by local bytearray objects. Once the helpers returned, those local variables were freed, leaving the C library with dangling pointers when addWrites()/check() later read the keys. Store the backing bytearray on the returned WriteRange/ReadRange objects as private `_begin_buf` / `_end_buf` attributes. Python keeps them alive for the lifetime of the range object, so the C pointer is always valid. Closes #42	2026-06-22 14:09:18 -04:00
weaselbotandandrew	ff0722728a	Include <sys/syscall.h> in ServerBench.cpp for SYS_perf_event_open ... `ServerBench.cpp` calls `syscall(SYS_perf_event_open, ...)` but did not include `<sys/syscall.h>`, relying on `<unistd.h>` to transitively provide the `SYS_*` constants. On toolchains where that does not happen, the build fails with `SYS_perf_event_open` not declared. Add the missing include so `server_bench` compiles on any platform that provides `perf_event_open`. Resolves weaselab/conflict-set#41.	2026-06-22 14:09:14 -04:00
weaselbotandandrew	60881419b8	Assert return values in test_update_zero_should_commit / conflict ... Add the missing [Result.COMMIT] / [Result.CONFLICT] assertions so that the two regression tests actually verify the expected conflict-set outcome, not just that the calls do not crash.	2026-06-22 14:09:10 -04:00
weaselbotandandrew	4515af3662	CMake: add target-level dependencies for conflict-set-object ... The custom command that links conflict-set.o depends on $<TARGET_OBJECTS:conflict-set-object>, but that generator expression does not create a target-level dependency. With the Unix Makefiles generator, parallel builds can start building the consuming libraries before the object library's build rule is available, producing: gmake[3]: *** No rule to make target 'CMakeFiles/conflict-set-object.dir/ConflictSet.cpp.o', needed by 'conflict-set.o'. Stop. Add add_dependencies() so that conflict-set and conflict-set-static cannot build until conflict-set-object has produced its object files. Closes #47	2026-06-22 14:09:04 -04:00
weaselbotandandrew	4dc5f7f75c	Fix self-move-assignment and leak in ConflictSet move-assignment ... The user-declared move-assignment operator overwrote `impl` without first destroying the existing implementation object, leaking all memory and resources owned by the left-hand side. Self-move-assignment also set `impl` to nullptr, leaving the object invalid and leaking the old state. Fix all three implementations (ConflictSet.cpp, SkipList.cpp, HashTable.cpp) to guard against self-assignment and to destroy/free the old `impl` before taking ownership of `other.impl`.	2026-06-22 14:08:59 -04:00
weaselbot	7eaac2a184	Make ConflictSet non-copyable in C++98/C++03 ... `ConflictSet(const ConflictSet&)` and `operator=(const ConflictSet&)` were only deleted for C++11 and later. In C++98/C++03 the compiler implicitly generated public copy operations, so copying a ConflictSet shared the opaque `Impl*` and caused a double-free on destruction. Declare both operations private and leave them undefined when `__cplusplus <= 199711L`, matching the standard pre-C++11 idiom for move-only types. Guard the declarations with `defined(__cplusplus)` so they are not exposed to C90 compilation units. Closes #48	2026-06-22 13:43:04 -04:00
weaselbotandandrew	742d920aa1	Set restype=None for void-returning C functions in conflict_set.py ... ConflictSet_check, ConflictSet_addWrites, ConflictSet_setOldestVersion, and ConflictSet_destroy return void in C, but the Python wrapper left their ctypes restype at the default c_int. Set restype = None for each to match the C API contract and avoid undefined behavior from reading the return register of void functions.	2026-06-21 19:25:27 -04:00